Understanding ETNs on VIX Futures

Electronic copy available at: http://ssrn.com/abstract=2043061

Understanding ETNs on VIX Futures

Carol Alexander and Dimitris Korovilas

ICMA Centre, Henley Business School at Reading

This Version: May 19, 2012

ABSTRACT

This paper aims to improve transparency in the market for direct, leveraged and inverse exchange-traded notes (ETNs) on VIX futures. The first VIX futures ETNs were issued in 2009. Nowthere are about 30 of them, with a market cap of about $3 billion and trading volume on some ofthese products can reach $5 billion per day. Yet volatility trading is highly complex and regulatorsshould be concerned that many market participants lack sufficient understanding of the risks theyare taking. We recommend that exchanges, market-makers, issuers and potential investors, as wellas regulators, read this paper to improve their understanding of these ETNs.

We provide a detailed explanation of the roll yield and convexity effects that drive the returnson VIX futures ETNs, and we track their volatility and assess their performance over an eight-yearperiod starting in March 2004, by replicating their values using daily close VIX futures prices. Weexplain how ETN issuers can construct almost perfect hedges of their suite of ETNs and control theirissue (most ETNs are callable) to make very significant profits under all bootstrapped scenarios.However, market knowledge has precipitated front-running of the issuers hedging activities, makingprofits more difficult to control. Moreover, for hedging the ETNs such large positions must be takenon VIX futures that the ETN market now leads the VIX futures that they are supposed to track.The result has been an evident increase in the volatility of VIX futures since 2009. If this increasein statistical volatility induces an increase in VIX futures implied volatility, a knock-on effect wouldbe higher prices of VIX options whilst S&P options are unaffected.

A previous discussion paper, Alexander and Korovilas (2012), provided incontrovertible evi-dence that single positions on direct VIX futures ETNs of any maturity including mid-term andlonger-term trackers could only provide a diversification/hedge of equity exposure during the firstfew months of a great crisis of similar magnitude to the banking collapse in late 2008. By contrast,the present discussion paper shows that some highly attractive long-term investment vehicles canbe simply constructed by holding certain portfolios of VIX futures ETNs. In particular, we intro-duce a new class of roll-yield arbitrage ETN portfolios which we call ETN2 (because they allocatebetween direct and inverse VIX futures tracker ETNs) and ETN3 portfolios (that allocate betweenstatic and dynamic ETN2). These portfolios have positive exposure to mid-term direct-trackerETNs and (typically) negative exposure to short-term direct-tracker ETNs (equivalently, positiveexposure to short-term inverse-tracker ETNs). Their unique risk and return characteristics makethem highly attractive long-term investments, as well as superb diversifiers of stocks, bonds andcommodities.

Email: [email protected] and [email protected]

ICMA Centre, University of Reading, Reading, RG6 6BA, UK

Tel: +44 (0)118 378 6431

Electronic copy available at: http://ssrn.com/abstract=2043061

Contents

1 Introduction 1

2 Constant-Maturity VIX Futures 3

2.1 Constant-Maturity Prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.2 Investable Returns and the S&P Indices . . . . . . . . . . . . . . . . . . . . . . . . . 52.3 Roll Cost and Convexity Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3 Statistical Analysis of S&P Indices 9

3.1 Volatility Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93.2 Correlation Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123.3 Implications for Roll-Yield Arbitrage Trades . . . . . . . . . . . . . . . . . . . . . . . 13

4 Taxonomy of VIX Futures ETNs 14

4.1 The XVIX and XVZ Exchange-Traded Notes . . . . . . . . . . . . . . . . . . . . . . 154.2 Replication Results for Volatility ETN2 . . . . . . . . . . . . . . . . . . . . . . . . . 174.3 Introducing Volatility ETN3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184.4 Diversifying Risk along the VIX Futures Term Structure . . . . . . . . . . . . . . . . 20

5 Performance of ETNs: The Investors Perspective 21

5.1 Performance Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215.2 Empirical Performance of ETNs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225.3 The Minimum-Variance ETN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235.4 Risk and Return on ETN2 and ETN3 . . . . . . . . . . . . . . . . . . . . . . . . . . 255.5 Comparison with Standard Asset Classes . . . . . . . . . . . . . . . . . . . . . . . . 26

6 Hedging ETNs: Concerns for Issuers and Regulators 28

6.1 Perfect Hedges Based on Indicative Values . . . . . . . . . . . . . . . . . . . . . . . . 286.2 Empirical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296.3 Scenario Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306.4 The Regulators Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

7 Summary and Conclusions 34

References 38

Appendix 38

1. Introduction

Futures contracts on the Standard and Poors (S&P) 500 volatility index (VIX) began trading

on the CBOE1 futures exchange in March 2004. Because the VIX index is not tradable there is

no unique closed-form, arbitrage free, cost-of-carry relationship connecting the VIX index with

the price of a VIX futures. In fact, there is often a sizeable difference between the index and its

futures prices.2 Still, the futures price represents the risk-neutral expectation of VIX at maturity,

and as such VIX futures offer a volatility exposure that is still very highly correlated with the VIX

index and with the over-the-counter (OTC) S&P 500 variance swaps brokered by investment banks.

Moreover, unlike variance swaps, the futures have no credit risk. Hence, the investment side of a

commercial bank no longer needs to rely solely on risky OTC trades to gain exposure to volatility.

Since 2004, VIX futures have been actively promoted as having diversification benefits and

other unique characteristics. Sophisticated market players now trade VIX futures for speculation,

directional exposure, arbitrage, diversification and vega hedging. And now any type of investor

e.g., a pension funds or an individual has easy access to volatility trades on the New York

stock exchange (NYSE) through the exchange-traded notes (ETNs) that track constant-maturity

VIX futures. These products have some adverse features not shared by futures: (i) They retain

the credit risk of the issuer, which has been relatively high since the banking crisis; (ii) A small

investor may be trapped into an illiquid investment because the issuer will only redeem the shares

early in large lots; (iii) Many ETNs have a callable feature whereby the issuer can call back the

shares at any time, with a short call notice period. On the other hand, ETN issuers charge only a

small early redemption fee and an annual service fee related to their hedging costs.

In 2009, Barclays Bank PLC issued VXX and VXZ, their 1-month and 5-month constant-

maturity VIX futures trackers. Their performance is directly linked to that of the S&P 500 VIX

Short-Term Futures Index and the S&P 500 VIX Mid-Term Futures Index respectively.3 More

brokers, notably ETRACS of UBS AG and VelocityShares of Credit Suisse, quickly followed suit

with other tracker ETNs, 2 leveraged products and inverse exposures to the S&P constant-maturity VIX futures indices. By December 2011 about 30 VIX-linked ETNs were trading in

very high volumes on secondary markets,4 the primary market being the NYSE Arca. About

$875 million was traded per day, on average, during the first two months of 2012 (not a particularly

volatile period) on two of these ETNs; VXX, the Barclays iPath 1-month constant maturity tracker

and TVIX, its supra-speculative, twice leveraged extension issued by Credit Suisse in November

2010. The U.S. Securities and Exchange Commission (SEC) are currently scrutinizing the dangers

of ever-more complex leveraged exchange-traded instruments. 5 After many adverse press reports

about TVIX, Bloomberg reported that the SEC investigators will review the product.6

1Chicago Board Options Exchange.2Another (minor) difference is that VIX futures are settled on the special opening quotation price, which is based

on traded prices of S&P 500 options fed in the VIX formula, whereas the VIX itself is based on the options mid price.3See Standard & Poors (2011b) for details of the methodology underpinning their calculation.4Including four ProShares volatility-based products that trade as exchange-traded funds.5See Reuters, February 23, 2012: Analysis: Out of control? Volatility ETN triggers risk concerns.6See: SEC Said to Review Credit Suisse VIX Note. Bloomberg, March 29, 2012.

1

Issuers of ETNs, the exchanges that list volatility products, and S&P (which calculates the

indices used for indicative values) are all promoting VIX futures and their ETNs as products

that are suitable for investors seeking diversification, such as pension funds and mutual funds.

However, the frenetic ETN market activity has spilled over to the VIX futures market, increasing

the volatility of VIX futures so that they have now become some of the most risky of all exchange-

traded instruments.7 Besides, Alexander and Korovilas (2012) provide a detailed and thorough

demonstration that individual position in VIX futures, or their ETNs, offer no opportunities for

diversification of equity exposure, except during the onset of a major crisis. In short, they should

be entered only as speculative trades.

On the other hand, one of the most important conclusions of this paper is that certain portfolios

of VIX futures, or their ETNs, which typically take a short position on short-term VIX futures

and a long position on longer-term VIX futures, can offer unique risk and return characteristics

that should be attractive to many long-term investors, as well as providing superb opportunities

for diversification of stocks, bonds and/or commodities. These portfolios might be referred to as

a new class of roll-yield arbitrage ETN portfolios, although they are not riskless. The idea is to

trade on the differential roll costs at different points along the VIX futures term structure. We call

these portfolios ETN2 when they allocate between direct and inverse VIX futures tracker ETNs,

and ETN3 portfolios when they allocate between static and dynamic ETN2.

We explain the mechanics under-pinning the design of two recently-launched volatility ETNs

the XVIX, issued by UBS AG in November 2010, and the XVZ, issued by Barclays Bank PLC in

August 2011 that fall into the category of ETN2. Different ETN2 products can have quite diverse

risk and return characteristics. For instance, by replicating the indicative values of the XVIX and

VXZ daily, from inception of VIX futures in March 2004 until December 2011, we show that: (i)

the XVIX has almost zero correlation with the VIX, a relatively low volatility and it performs best

during tranquil, trending market conditions; and (ii) the XVZ has a strong positive correlation

with VIX, a relatively high volatility, and is highly profitable only during volatile periods.

This finding motivates a new class of ETN3 portfolios that allocate between ETN2s.8 These

ETN3 have superior performance and greater diversification potential than the corresponding static

and dynamic ETN2s, according to a wide variety of performance criteria. We label these static and

dynamic ETN3 portfolios CVIX and CVZ, respectively.

ETN issuers can hedge their exposure to early redemption of ETNs perfectly, provided they can

trade VIX futures at the daily closing price, because it is this price that determines the redemption

value for a VIX futures ETN. Given the service fee charged we explain how issuers should guarantee

significant profits, net of hedging costs, provided they issue a controlled portfolio of ETNs. However,

there has recently been large-scale front-running activity on the very high volumes that must be

traded near the market close on prompt and second to mature VIX futures for the purpose of

7For instance, during the onset of the Eurozone debt crisis in August 2011 the statistical (GARCH) volatility ofthe prompt VIX futures contract exceeded 200%, at a time when the average value traded on this contract aloneexceeded one billion USD per day.

8Like all other volatility ETN innovations, ETN3 can be replicated by trading a term structure of VIX futures.

2

hedging the two largest ETNs, VXX and TVIX. As a result, Credit Suisse stopped the issue of

TVIX (sending its traded price to a 90% premium over indicative value at one point, such is the

speculative demand on this product) and re-opened it only on the condition that the hedging risk

be passed on to the market makers, i.e. they guarantee VIX futures closing prices to Credit-Suisse.

The outline of this discussion paper is as follows: Section 2 begins with explaining the fundamen-

tal building-blocks for constructing volatility ETNs, viz. indices of investable, constant-maturity

VIX futures which are computed by the S&P. These indices determine the indicative value of VIX

futures ETNs. Section 3 analyses the statistical features of constant-maturity VIX futures tracker

products, and their implications for new volatility ETNs that are based on trading differential roll-

yield effects along the term structure. Section 4 describes how the ETN2 and ETN3 products that

we consider could be replicated using the term structure of VIX futures, and then examines their

performance using a variety of criteria over the eight-year sample period. In Section 5 we explore

the risk and return characteristics of one static and one dynamic ETN3, comparing their perfor-

mance with the ETNs already available, and demonstrating their great potential for diversification

of exposures to other traditional asset classes: equities, commodities and bonds. Section 6 explains

how the issuers of ETNs can structure their products and the size of their issue so that they can

hedge their liabilities almost perfectly, and moreover make significant profits, provided the hedge

on VIX futures is traded at the closing price on the CBOE. Section 7 concludes.

2. Constant-Maturity VIX Futures

For the period March 26, 2004 to December 31, 2011 we obtained Bloomberg data on the daily

close (last traded price or index value) for: (i) all VIX futures contracts; (ii) VIX and VXV, i.e. the

30-day and 93-day implied volatility indices calculated by CBOE; (iii) the S&P constant-maturity

VIX futures indices; (iv) the two most well-established volatility ETNs, i.e. the VXX and VXZ,

Barclays 1-month and 5-month constant-maturity VIX futures trackers; (v) XVIX and XVZ, the

two recently-launched ETN2.9

2.1. Constant-Maturity Prices

From the VIX futures prices we construct two sets of six synthetic time series, each set repre-

senting constant maturities of m = 30, 60, . . . , 180 days. The first set is a non-investable futures

price time series and the second is a futures return time series, which is investable. The constant-

maturity futures price on day t is derived as:10

pmt = tpst + (1 t)plt, t =

l ml s (1)

9(i) are only available from the inception of VIX futures on March 26, 2004; (iii) start only in December 2005 andprior to this we construct the indices by applying the S&P methodology to traded VIX futures; (iv) are availableonly since their launch in January 2009, but we replicate their indicative values using actual and reconstructed S&Pindices; (v) have very few traded prices, starting only in November 2010 and August 2011 respectively, but again wereplicate their indicative values as for (iv).

10Note that in Eq. (1) and also in Eq. (2) below, we exclude observations with maturity less than five businessdays. This is to avoid irregularities caused by near-to-maturity trading, as well as the special settlement process inthe VIX futures market.

3

where pm is the synthetic futures price for a specific maturity m, ps and pl are the prices of the

shorter and longer maturity exchange-traded futures contracts that straddle the maturity m, with

s < m < l, and each maturity is measured in calendar days to expiry. These price series provide a

visualization of the futures term structure and its properties, based on market traded prices but,

returns based on these synthetic prices are not realizable.

Jan05 Jan06 Jan07 Jan08 Jan09 Jan10 Jan11 Jan120

10

20

30

40

50

60

70

80

90

Volat

ility (%

)

VIX 30day 60day 90day 120day 150day 180day

Fig. 1 Term structure evolution of the six constant-maturity VIX futures series, and the VIX 30-day impliedvolatility index. March 2004 December 2011.

Fig. 1 depicts the VIX futures price series for the six different maturities under consideration,

and the VIX 30-day implied volatility index that underlies the futures. Since the financial crisis

starting in mid 2007 volatility has never returned to its previous levels. In September 2008, pre-

cipitated by the Lehman Brothers collapse, VIX futures were trading at around 50%. Since then

volatility was especially high during the Greek crisis of May 2010 and the wider Eurozone debt

crisis beginning in August 2011.

The term structure of VIX futures is typically in contango. Brief periods of backwardation

only accompany excessively volatile periods. For our later analysis (in section 4.3) it is interesting

to note that, over the entire sample, the 30-day futures price was greater than (less than) the

150-day futures price about 24% (76%) of the time. In some periods, particularly from Q2-2007

to Q2-2008, the long-term contracts at 150 or 180 days to expiry are trading at very similar levels

to the short-term contracts at 30 or 60 days to expiry, which implies that the market believes the

VIX is at its long-term level.

When the term structure is in backwardation its slope is typically inconsistent with the time-

series properties of VIX. For instance, on January 2, 2009 the (synthetic) 180-day futures closed

at 36%; but this overestimated the implied value of VIX on July 2, indeed on that date the VIX

was below 30%. In other words, the VIX mean-reverts more quickly than the term-structure of

4

VIX futures would imply. Even if the slope is underestimated during periods of backwardation, the

slope of the term structure is still very steep. It has been particularly pronounced since the onset

of the banking crisis in 2008. This feature makes VIX futures particularly suitable for the type of

differential roll-yield trades that we shall propose in this paper.

30 60 90 120 150 18010

20

30

40

50

60

70

VIX Fu

tures

Prices

Days to maturity

06/03/201213/03/201213/11/200820/11/200828/11/2008

Fig. 2 Term structure of VIX futures while in contango (2010) and backwardation (2008).

Fig. 2 depicts how the typical shift and slope of the VIX futures term structure varies, according

to whether the market is in contango or backwardation. In November 2010 the term structure

exhibited contango, being steeper at the short-end than at the longer maturities. Also, the shifts

in the term-structure are relatively small each curve is depicted at weekly intervals. By contrast,

during November 2008 the market was in distinct backwardation. At such times the (now negative)

slope of the term structure is very much steeper at the short-end, and the shifts are of greater

magnitude. For instance, the term structure shifted upward by 17 volatility points at the short-end

during a single week, and then returned almost to the previous level by the end of the next week.

2.2. Investable Returns and the S&P Indices

The indicative values of VIX futures ETNs are based on the S&P indices of investable constant-

maturity (ICM) VIX futures derived from their daily closing prices. As explained by Galai (1979),

an ICM discretely compounded return on VIX futures may be obtained via linear interpolation

between the returns on the two futures with maturity dates either side of the constant maturity.

To construct such a series we set

rmt+1 = trst + (1 t)rlt+1, t = tpst/pmt , rit+1 =

pit+1 pitpit

, (2)

5

where rit+1 are the individual discretely-compounded returns on the tradable futures contracts with

maturities i = s (short) and i = l (long).11 Basing an ETN on the ICM returns given by Eq. (2)

retains the one-to-one correspondence between the weights derived for the ETN and the weights

on the futures that are actually traded. In other words, the ETN issuer may replicate the product

using the exchange-listed VIX futures, because at any point in time there is a unique interpolation

constant t for each ICM futures return, which is used via Eq. (2) to distribute into weights on

traded VIX futures.12

Jan05 Jan06 Jan07 Jan08 Jan09 Jan10 Jan11 Jan12$0

$20

$40

$60

$80

$100

$120

$140

Portf

olio V

alue

30day 60day 90day 120day 150day 180day

Fig. 3 Evolution of the value of $100 invested in the VIX futures with constant-maturity returns, constructed usingEq. (2), in March 2004. These represent the S&P indices of maturity T , with T = 30, 60, . . . , 180 days.

To illustrate the difference between the constant-maturity price series Eq. (1), which are dis-

played in Fig. 1 and the attainable returns for an investor following the constant-maturity method-

ology as in Eq. (2), Fig. 3 depicts the performance of a theoretical $100 investment in each of the

six constant maturities. The series represent the S&P indices of maturity 30, 60, . . . , 180 days.

To attain the ICM return one must rebalance the portfolio of the two straddling VIX futures

daily to keep the required constant-maturity exposure. Since the VIX futures term structure is

almost always in contango, at each rebalancing there is a small but almost always negative roll

cost created by selling the lower price shorter-term futures and buying the higher price longer-

term futures. This small daily roll cost creates highly negative long-run returns for the investor

in a short-term VIX futures tracker ETN. But because of the convexity in the VIX futures term

structure, replicating a long-maturity ETN performs relatively well, because the difference in roll

11Whenever there are not two contracts that straddle the desired maturity, linear extrapolation rather than linearinterpolation is performed using the two closest contracts to that maturity. In that case a short position in one ofthe two contracts used in the calculation is necessary.

12Note that returns on VIX futures, as for any other futures contract, directly represent an excess over the currentrisk-free rate return see Bodie and Rosansky (1980) and Fortenbery and Hauser (1990) for further explanation.

6

costs between the two straddling VIX futures contracts is smaller than it is at the short end of the

term stucture. That is, the size of the roll yield depends on the slope of the term structure, which

is steepest at the short end. Hence, the negative roll yield when the term structure is in contango is

much greater at the short end of the VIX futures term structure. This is the main reason why the

1-month tracker VXX has performed so much worse than the 5-month tracker VXZ. For instance,

$100 invested in the VXZ on January 29, 2010 was worth a little over $60 by December 30, 2011,

but the same invested in VXX was worth less than $9!

2.3. Roll Cost and Convexity Effects

The discretely-compounded daily return derived from the price index pmt is not realisable, and

not equal to rmt . The difference between them is called the roll yield of maturity m and its negative,

the roll cost, is denoted cmt . The roll cost captures the loss (or gain) from rolling a futures position

from a shorter maturity contract to one of a longer maturity when the futures market is in contango

(or backwardation) on day t.

If we ignore transactions costs so that all prices are mid prices, we can simply define the roll

cost to be Eq. (2) minus the one-period return based on Eq. (1). On applying a little algebra it

can be shown that:13

cmt+1 = rmt+1

[(pmt+1 pmt )/pmt

]=

pst+1 plt+1pmt

(t+1 t). (3)

Typically (i.e. unless a new pair of contracts is rolled into) we have t > t+1. For instance, when

rebalancing inter-week rather than over a weekend, t t+1 = (l s)1.14 When the market isin contango, plt > p

st so the roll cost (3) is positive. The roll cost is only negative (i.e. roll yield

positive) when the term structure is in backwardation. The VIX futures term structure is very often

in contango, as we have seen from Fig. 1, so a very large and negative roll yield has substantially

eroded the returns realised on all the standard tracker ETNs products since their launch.

Convexity of the term structure induces a different sensitivity of ETNs returns during periods of

contango and backwardation. To see this, note that the short end of the VIX futures term structure

is much more volatile, i.e. rst is usually of much greater magnitude than rlt. The sensitivity of r

mt

to rst and rlt depends on whether term structure is in backwardation or contango. For instance, set

t = 0.5, so that t = (2pmt )

1pst and 1t = (2pmt )1plt. In contango, t < 1t and the constant-maturity return rmt+1 given by (2) has a relatively low sensitivity to r

st+1. But in backwardation,

where returns typically have much greater magnitude (cf. Fig. 2) we have t > 1 t so rmt+1 hasa higher sensitivity to rst+1. This convexity effect means that, as the market swings from contango

into steep backwardation at the beginning of a crisis, very rapid gains are made on the synthetic

short-term VIX futures; but thereafter equally rapid losses are made, as the backwardation declines

and the term structure returns to contango.

13The S&P uses another definition but without apparent justification, e.g. Standard & Poors (2011a).14The level of l s depends on the distance between the two contracts used in the constant-maturity calculation,

which is most often either 28 or 35 days in our sample.

7

Jan07 Jan08 Jan09 Jan10 Jan110

50000

100,000

150,000

200,000

250,000

300,000In

dex L

evel

1month 5month Inverse 1month 2 x 1month

Fig. 4 S&P 30-day and 150-day constant maturity indices, the inverse 30-day index and 2 leverage on the 30-dayindex, December 2005 December 2011.

To witness this convexity effect in action, Fig. 4 displays the S&P 500 constant-maturity indices

underlying four major ETNs: the VXX. VXZ, XIV and TVIX.15 These indices were first quoted

in December 2005, starting at a value of 100,000.16 The convexity effect induces a heightened

sensitivity to the short-term return during a period of strong backwardation such as the banking

crisis at the end of 2008. Thus, as the market swung from contango to backwardation at the

beginning of the crisis, the direct tracker indices jumped upward. For instance, during the two

months following the collapse of Lehman Bros on September 17, 2008 the 30-day index rose by

more than 200%, the inverse 30-day index fell by 73%, the 2 leveraged 30-day index rose by670%, and the 150-day index rose by 73%. But as the market swung back to contango the indices

lost the value they gained and the higher the gain the greater and more rapid the subsequent losses.

The convexity effect is positive for purchasers of direct tracker ETNs, but leads to large negative

drawdowns for the inverse tracker products. For instance, the XIV, whose indicative value is

represented by the red line in Fig. 4, lost about 50% of its value during the first few days of the

Eurozone crisis in August 2011.

We conclude that the direct, inverse and leveraged tracker ETNs are likely to be amongst the

most risky of all exchange-traded products. We now move to a statistical analysis of the returns

on investable constant-maturity VIX futures, which will verify this claim.

15Daily returns on VXX and VXZ are returns on the S&P 1-month and 5-month constant-maturity total returnindices, respectively, less the investor fee. Credit Suisse issued the 2 leveraged product TVIX (and the XIV inverseETN whose returns are minus the returns on the S&P 30-day index, less a 1.35% annual fee) in November 2010.

16Prior to this date their evolution is proportional to the series displayed in Fig. 3, which we started at 100. It isinteresting to note that between March 2004 and December 2005 the 30-day S&P index lost over 80% of its value.

8

3. Statistical Analysis of S&P Indices

Panel A of Table 1 reports summary statistics on the six ICM VIX futures returns that de-

termines the value of the S&P indices, computed using Eq. (2) with maturities, 30, 60, . . . , 180

calendar days.17 As we move to longer maturities the returns represent longer term expectations

of volatility and are consequently less variable over time. This is consistent with the well-known

Samuelson effect in all futures markets see Samuelson (1965). However, the decrease in volatility

as we move along the term structure of VIX futures is much more pronounced than it is in most

other financial and commodity futures. Another unique and prominent feature is that, due to the

roll-cost effects explained in the previous section, there are much larger negative returns at the

short end of the term structure than at the long end. Panel B presents the correlation matrix of

these returns over the entire sample. Clearly the series form a very highly correlated system, with

correlations ranging between 0.86 and 0.98.

Table 1

Descriptive statistics for returns on investable, constant-maturity VIX futures.

Panel A: Summary statisticsContract Annualized Mean Volatility Sharpe Ratio Skewness Kurtosis Total Return

30-day -33.57% 57.20% -0.5869 1.16 5.38 -97.94%60-day -16.92% 43.35% -0.3902 0.88 4.46 -87.12%90-day -4.79% 36.73% -0.1305 0.77 4.22 -59.24%120-day -2.15% 32.84% -0.0655 0.72 4.40 -44.37%150-day -0.36% 30.29% -0.0119 0.67 4.52 -31.90%180-day 0.62% 28.69% 0.0215 0.68 4.75 -23.76%

Panel B: Correlations30-day 60-day 90-day 120-day 150-day 180-day

30-day 1.0060-day 0.97 1.0090-day 0.93 0.98 1.00120-day 0.91 0.95 0.98 1.00150-day 0.88 0.92 0.95 0.98 1.00180-day 0.86 0.90 0.93 0.95 0.98 1.00

Panel A reports the summary statistics for investable, constant-maturity VIX futures daily returns. Sample means,standard deviations and corresponding Sharpe ratios are annualized based on 250 trading days per year. We presentfigures for excess kurtosis not kurtosis here and later, in Tables 6 and 9. Panel B is the correlation matrix for thereturns under consideration. All figures are significant at the 1% level. The sample is between March 26, 2004 andDecember 30, 2011.

3.1. Volatility Analysis

We examine the statistical volatility of the indicative values for VIX futures ETNs by applying a

generalized autoregressive conditional heteroscedasticity (GARCH) framework to daily returns on

S&P indices of different maturities. To capture the possibility of an asymmetric volatility response

to shocks we employ the well-known Glosten, Jagannathan, and Runkle (1993) (GJR) model. That

17The results here omit transaction costs and the investment fee for VIX futures ETNs (these are analysed later,in Sections 6 and 5).

9

is, the return rt is assumed to follow the conditional mean and variance equations:

rt = c+ tzt, zt NID(0, 1), (4)2t = +

2t1 + 1{t1

Table 3

Maximum likelihood estimation of GJR models on constant-maturity VIX futures returns.

30-day 60-day 90-day 120-day 150-day 180-day

Panel A: March 26, 2004 - December 31, 2010 2.94 105 1.58 105 6.47 106 4.12 106 3.24 106 3.61 106

(8.09) (6.52) (6.57) (4.81) (5.65) (7.02)

0.1878 0.1459 0.1360 0.1463 0.1377 0.1343(11.16) (9.49) (9.46) (9.80) (10.51) (10.60)

0.8602 0.8854 0.8932 0.8842 0.8896 0.8877(68.39) (82.82) (89.57) (84.39) (95.76) (99.78)

-0.1510 -0.1187 -0.0886 -0.0795 -0.0698 -0.0661(-8.49) (-6.89) (-5.90) (-5.09) (-4.85) (-4.83)

+ 0.5 0.1123 0.0866 0.0917 0.1065 0.1028 0.1013

51.67% 37.62% 32.66% 33.37% 32.67% 28.62%

Panel B: January 1, 2011 - March 31, 2012 1.21 104 7.57 105 3.98 105 3.00 105 2.44 105 1.96 105

(2.25) (3.09) (2.52) (2.27) (2.34) (2.08)

0.1365 0.2044 0.2144 0.1906 0.1762 0.1653(3.41) (3.61) (3.29) (3.01) (2.73) (2.50)

0.8393 0.7721 0.7666 0.7768 0.7805 0.7888(17.04) (15.27) (16.55) (16.54) (15.69) (14.84)

-0.0544 -0.0417 -0.0083 0.0127 0.0370 0.0453(-1.09) (-0.54) (-0.10) (0.15) (0.41) (0.48)

+ 0.5 0.1093 0.1835 0.2102 0.1969 0.1947 0.1880

76.70% 65.33% 65.55% 53.51% 49.54% 46.00%

In-sample maximum likelihood estimation of GJR-GARCH coefficients for each VIX futures series. Numbers inparentheses denote tstatistics under the null hypothesis that coefficients are equal to zero. All coefficients aresignificant at the 1% level. denotes the long term average for volatility, i.e. for the GJR model the annualizedsquared root of 2 = /(1 0.5).

the long-term volatility of 30-day futures, at 56.27%. Moreover, the long-term volatility does not

fall much further for the 90-day and longer maturity contracts: all longer-maturity contracts have

similar . The implication here is that equity investors expect VIX to display a very rapid mean-

reversion, typically taking more than 1 month but less than 2 months to revert to normal levels

after a shock. This is also evident from historical volatilities presented in Table 1.

Fig. 5 depicts the GJR conditional volatility estimates over the entire sample. The conditional

volatility is much higher and more variable for the 30-day futures compared with other maturities.

A maximum volatility of more than 200% for the 30-day VIX futures occurred during the onset

of the Eurozone debt crisis of August 2011. The TVIX was launched one month later, but had it

existed in August 2011 its volatility would have been over 400%! The volatility of VIX futures also

peaked at the time of the Lehmann Brothers collapse in September 2008, and many times after

this. Indeed all VIX futures, even those of longer maturities, have displayed numerous jumps in

their volatilities along with a generally increasing level of volatility.

As the trading volume on VIX futures has increased, due to large-scale hedging activity of

ETNs issuers, they have become more volatile relative to the VIX index itself. For instance,

between January 2, 2007 and March 31, 2010, a period when the average VIX index value was

27%, the 30-day VIX futures had an average volatility of 60%. But between April 1, 2010 and

11

Jan05 Jan06 Jan07 Jan08 Jan09 Jan10 Jan11 Jan120

20%

40%

60%

80%

100%

120%

140%

160%

180%

200%VI

X Fu

ture

s Vola

tility


Fig. 5 GJR-GARCH volatilities for the six VIX futures series based on the maximum likelihood estimation of themodel for conditional variance: 2t = +

2t1 + 1{t1 0. The mth component is determined by the eigenvector belonging tothe mth largest eigenvalue, so that the proportion of total variation explained by the mth principal

component is m/(1+ +n). This way we identify how many key factors explain most of the totalrisk in the system. Note that principal components are, by construction, mutually uncorrelated

because W is an orthogonal matrix.

Table 4 reports the PCA of the correlation matrix of S&P index returns, estimated over the

entire sample. Panel A presents the proportion of total variation in the system explained by each

component. The first component alone explains almost 95% of the movements in the system and

12

Table 4

Principal components analysis on the correlation matrix of VIX futures constant maturity returns.

Contract PC1 PC2 PC3 PC4 PC5 PC6

Panel A: Proportion of variation explained94.73% 3.44% 1.04% 0.49% 0.19% 0.11%

Panel B: Eigenvectors30-day 0.3980 -0.6185 0.5062 -0.3890 0.2166 0.067460-day 0.4101 -0.3886 -0.1205 0.5116 -0.5848 -0.250090-day 0.4143 -0.0962 -0.5043 0.2797 0.4912 0.4953120-day 0.4138 0.1774 -0.4293 -0.4517 0.1330 -0.6256150-day 0.4105 0.4019 0.0443 -0.3661 -0.5256 0.5077180-day 0.4026 0.5140 0.5373 0.4130 0.2754 -0.1963

Panel A reports the proportion of total risk explained by each factor driving the term structure of the S&P indexreturns, based on the sample from March 2004 to December 2011. Panel B reports the eigenvectors (i.e. factor weightsfor the principal component representation) for each return series.

with three factors representing the system we are able to explain over 99% of the total variation.

Panel B of Table 4 reports the eigenvectors, i.e. the factor weights for each S&P index return series.

Reading across the first row in panel B and using only three components, we have with over 99%

accuracy, the following representation of the standardized returns r30t on 30-day S&P index:19

r30t 0.3980PC1,t 0.6185PC2,t + 0.5062PC3,t. (6)

Similar three-factor representations for futures of other maturities show that the first component

captures an almost exact parallel shift in the term structure, in that if PC1 moves by 1% and the

other components remain fixed, then all standardized futures shift upwards by about 0.4%. The

first eigenvalue tells us that 94.73% of the movements in standardized VIX futures over the sample

consist of parallel shifts of this type. The second component represents an almost exact linear

tilt in the standardized term structure, with shorter maturities increasing whilst longer maturities

decrease. Only 3.44% of the variation in the standardized VIX futures term structure can be

attributed to movements of this type.20

3.3. Implications for Roll-Yield Arbitrage Trades

The typical movements of the S&P indices may be decomposed into changes arising from: (a)

the roll cost cmt , representing a slide along the VIX term structure from a maturity of i days to

maturity of i 1 days,21 for each contract i = s, l straddling the maturity m; and (b) a movementof the entire VIX futures term structure from day t 1 to day t. The roll cost is relatively smalland highly predictable, compared with the large and unpredictable movements of type (b). A

19The standardized return r30t is the return at time t minus its sample mean divided by its sample standarddeviation.

20PCA is not merely a statistical tool for gaining further insight to the statistical behaviour of the term structure ofVIX futures. When applied to the VIX futures covariance matrix it becomes a tool for generating GARCH covariancematrices of VIX futures. In the appendix these covariance matrices will be used to examine how effectively the tradersin VIX futures can diversify their risk across the term structure.

21Or subtract more than 1 day over a weekend, when using business rather than calendar day counts. And heretypical excludes the effect when moving to a new pair of contracts.

13

roll-yield arbitrage seeks trades that are immunized against movements of type (b) so that their

main exposure is purely to the roll cost.22

The PCA analysis in the preceding sub-section shows that about 95% of the observed moves

in standardized returns are parallel shifts of the entire term structure. Hence, long-short trades in

standardized returns of different maturities should hedge about 95% of the large and unpredictable

daily movements of the term structure. That is, the amount gained or lost through the movement

at the short end of the term structure would be almost exactly offset by an equal opposite loss or

gain at the long end of the term structure.

A long-short ETN portfolio with positions weighted by the inverse of their respective volatilities

will capture this offsetting effect. For instance, since the 30-day futures have an average volatility

approximately twice that of the 150-day futures (cf. Table 1) a short position on 30-day VIX futures

(long position on XIV) offset by a long position of twice the size on 150-day futures (VXZ, VIIZ

or VXEE) should immunize against 95% the risk arising from movements of type (b). This type

of reasoning may well have been used by UBS to derive the XVIX, a roll-yield arbitrage ETN that

the bank issued in November 2010. The XVIX was the first example of an ETN2, having a static

allocation to the inverse 30-day VIX futures tracker and twice this allocation to the direct 150-day

VIX futures tracker.

4. Taxonomy of VIX Futures ETNs

Table 5 displays a chronology of VIX futures ETNs issued by Barclays (in red), Credit Suisse

(in blue) and UBS (in black). The first two columns give the constant maturity of the respective

indicative S&P index,23 and the maturity of the notes. The third column gives the leverage, i.e. 1

for a direct tracker, 1 for an inverse tracker, and 2 for a direct leveraged tracker. Then follows: theannual fee, higher for the inverse products because the issuer charges a significant fee for hedging

costs; the market cap of each note on February 29, 2012 and the average volume traded on each

note between January 2 and February 29, 2012.

The market cap on VXX is still almost double that of TVIX. Together their cap is 64% of the

total cap of all ETNs. Having been issued later, the UBS products have the lowest market cap (5%

of the total) and the Barclays products constitute 56% of the total market cap. The average daily

trading volume during the first two months of 2012 was 44.59 million shares over all ETNs, with

36.81 million shares traded on VXX and TVIX alone (amounting to about $875 million dollars per

day) and 23.87 million shares traded on average, per day, on just the Barclays products. On August

8, 2011 a total of $4,973,659,365 was traded on the VXX alone. Clearly, Barclays still dominates

the market although shares on the TVIX and XIV products of Credit Suisse are being traded in

increasingly high volumes.

The remainder of this section explores the potential benefits from roll-yield arbitrage trades

22We use the term arbitrage here because it has slipped into market usage. But, as mentioned in the introduction,it is not really an arbitrage because it is not riskless.

23And for the two ETN2, XVIX and XVZ, which attempt to exploit differential roll yields, the maturities of thetwo indices they allocate between.

14

Table 5

ETN chronology.

Maturity Service Market Cap. Average

Ticker* Inception VIX futures (m) Note (y) Leverage fee (million $) Volume

VXXa,b Jan-2009 1 10 1 0.89% 1,218.36 23,499,211VXZa Jan-2009 5 10 1 0.89% 233.26 329,478VIIXa Nov-2010 1 20 1 0.89% 15.82 32,605VIIZ Nov-2010 5 20 1 0.89% 7.63 829XVIX Nov-2010 1 vs. 5 30 1 0.85% 23.04 7,539XVZ Aug-2011 1 vs. 5 10 1 0.95% 187.33 25,816VXAA Sep-2011 1 30 1 0.85% 5.94 1,210VXBB Sep-2011 2 30 1 0.85% 7.19 370VXCC Sep-2011 3 30 1 0.85% 8.14 660VXDD Sep-2011 4 30 1 0.85% 8.37 192VXEE Sep-2011 5 30 1 0.85% 8.58 371VXFF Sep-2011 6 30 1 0.85% 9.20 182XXV Jul-2010 1 10 -1 0.89% 15.66 13,586XIVb Nov-2010 1 20 -1 1.35% 427.55 7,321,582

ZIVb Nov-2010 5 20 -1 1.35% 9.09 15,229AAVX Sep-2011 1 30 -1 5.35% 11.79 3,484BBVX Sep-2011 2 30 -1 5.35% 11.43 1,623CCVX Sep-2011 3 30 -1 5.35% 10.82 765DDVX Sep-2011 4 30 -1 5.35% 10.96 132EEVX Sep-2011 5 30 -1 5.35% 10.50 891FFVX Sep-2011 6 30 -1 5.35% 10.59 100IVOP Sep-2011 1 10 -1 0.89% 6.95 4,602VZZB Nov-2010 5 10 2 1.78%c 2.79 12,049TVIX Nov-2010 1 20 2 1.65% 676.04 13,315,361TVIZ Nov-2010 5 20 2 1.65% 7.32 4,464

We colour ETNs according to their issuer: Barclays Bank PLC (red), Credit Suisse AG (blue) and UBS AG(black). Except VXX, VXZ and XVZ, all other products have automatic termination clauses and all products exceptBarclays are callable at any time by the issuer. For products with leverage other than 1, actual leverage depends on day and time of transaction. All ETNs carry an extra redemption fee: 0.05% (red and blue) and 0.125% (black). Market cap as of February 29, 2012; average daily volume computed between January 2, 2012 and February 29,2012.a Options on these ETNs are available to trade on CBOE.b Split executed by the issuer: 1 for 4 (VXX), 10 for 1 (XIV), 8 for 1 (ZIV).c The 3m LIBOR is added on top of the service fee.

between inverse short-term and direct long-term VIX futures trackers. We begin by replicating and

analysing the performance of XVIX and XVZ, two such trading strategies that have recently been

issued as volatility ETNs. Then we introduce a new class of ETN3 products which trade on the

different performance of the ETN2 when the VIX futures term structure is in backwardation and

contango. This is followed by an analysis of more general diversifying trades along the entire term

structure, based on minimum variance portfolios.

4.1. The XVIX and XVZ Exchange-Traded Notes

The ETRACS Daily Long-Short VIX ETN (XVIX) is one of the thirteen VIX-based ETNs

brokered by UBS AG. It first traded on November 30, 2010 and matures on November 30, 2040.

15

It is a static exposure across the VIX futures term structure capturing 50% of the performance of

an inverse position in the VXX (the 30-day VIX futures tracker ETN) and 100% the performance

of a long position in the VXZ (the 150-day VIX futures tracker ETN). This strategy exploits the

roll yield differences between short-term and long-term contracts which are evident when the VIX

futures term structure is in contango, i.e. most of the time. Letting rxvixt denote the excess return

on the static carry strategy at time t we have:

rxvixt = r30t2

+ r150t (7)

where r30t and r150t are the excess returns earned on constant-maturity 30-day and 150-day VIX fu-

tures contracts respectively (i.e. the returns captured by the two respective S&P constant-maturity

indices as these are already described). Based on our previous comments in Section 3.3, we see that

this relationship hedges about 95% of the term structure movements so that it earns the differential

roll yield across the two maturities, which is typically positive because the market is in contango.

Barclays launched the dynamic VIX ETN (XVZ) in August 2011 and full details of this product

are given in Barclays Bank PLC (2011). Letting x30 and x150 denote the allocations on day t1 tothe 30-day and 150-day VIX futures contracts, respectively, the excess return on the XVZ strategy

at time t is given by:

rxvzt = x30t1r

30t + x

150t1r

150t (8)

The allocations x30 and x150 are derived relative to target allocations y30 and y150 using the following

process. Let the ratio of the 30-day implied volatility index (VIX) to the 93-day implied volatility

index (VXV) at time t be denoted Yt. Thus, when Yt < 1 (Yt 1) the short end of the VIX termstructure is in contango (backwardation). Now define target allocations [y30t , y

150t ] on 30-day and

150-day VIX futures as:

[y30t , y150t ] =

[0.3, 0.7] if Yt1 < 0.9,[0.2, 0.8] if 0.9 Yt1 < 1,

[0, 1] if 1 Yt1 < 1.05,[0.25, 0.75] if 1.05 Yt1 < 1.15,[0.5, 0.5] if Yt1 1.15.

(9)

The actual allocation applied on the strategy at time t is then:

x30t =

x30t1 if x30t1 = y

30t

min{x30t1 + 12.5%, y

30t

}if x30t1 < y

30t

max{x30t1 12.5%, y30t

}if x30t1 > y

30t

(10)

16

and

x150t =

x150t1 if x150t1 = y

150t

min{x150t1 + 12.5%, y

150t

}if x150t1 < y

150t

max{x150t1 12.5%, y150t

}if x150,t1 > y

150t

(11)

The XVZ is a relatively complex dynamic strategy that attempts to allocate between VXX and

VXZ to gain full advantage of the roll-yield differences along the VIX futures term structure, by

switching the positions in the short-term and long-term end of the term structure depending on

whether the VIX term structure is in contango or backwardation.

4.2. Replication Results for Volatility ETN2

We replicate the values of the XVIX and XVZ back to March 2004 using all available data on

VIX futures and S&Ps constant-maturity indices.24 Note that XVIX and XVZ are total return

products, i.e. the 91-day U.S. T-bill is added to their underlying index return. However, here we

present the excess returns on the products since, assuming they can borrow money at close to the

T-bill rate, it is excess returns that are of interest to large investors. We also deduct the annual

service fee (which is 0.85% and 0.95% per annum, for XVIX and XVZ respectively) so that the

results correspond to the actual excess return received by the investor. We also present results for

the two underlying ETNs, VXX and VXZ, after accounting for their service fee of 0.89% p.a..

Table 6

Descriptive statistics of ETNs and volatility roll-yield trades, and correlation matrix with VIX and VXV.

Panel A: Summary statisticsContract Annualized Mean Volatility Sharpe Ratio Skewness Kurtosis Total ReturnVXX -35.68% 56.38% -0.6328 1.08 5.03 -98.19%VXZ -1.50% 30.06% -0.0498 0.71 4.57 -37.36%XVIX 15.94% 13.42% 1.1875 -0.33 5.02 224.22%XVZ 17.75% 24.00% 0.7394 1.56 17.55 220.93%

Panel B: CorrelationsVXX VXZ XVIX XVZ VIX VXV

VXX 1.00VXZ 0.90 1.00XVIX -0.09 0.36 1.00XVZ 0.71 0.85 0.42 1.00VIX 0.87 0.79 -0.05 0.62 1.00VXV 0.89 0.82 -0.02 0.67 0.94 1.00

Panel A reports the summary statistics for the two volatility roll-yield trades under consideration (ETN2). Samplemeans, standard deviations and corresponding Sharpe ratios are annualized based on 250 trading days per year.Panel B is the correlation matrix for the ETN2 returns and the implied volatility indices, VIX and VXV. The sampleis between March 26, 2004 and December 30, 2011.

Table 6 presents some descriptive statistics based on our replicated sample. Compared with

24Since the S&P started quoting constant-maturity indices only in December 2005, the first 21 months of data arecomputed using our own computation of the constant-maturity futures returns based on Eq. (2). Note that betweenDecember 2005 and December 2011, the 30-day and 150-day returns based on Eq. (2) had correlation of 99.82% and99.32% with the Short Term and Mid Term S&P VIX futures indices respectively - not an exact 1 because of minordifferences of very few basis points arising from the difference in calendar/business days interpolation.

17

the tracker ETNs, the XVIX and XVZ are much less volatile. Moreover, they successfully turn the

negative volatility risk premium and negative roll-yield effect into positive returns. The simple,

static XVIX outperformed the more complex, dynamic XVZ on a risk-adjusted basis (as measured

by the Sharpe ratio) and although its average annual mean return was lower, its total return

was slightly greater than that of the XVZ. The higher volatility, large positive skew and very high

kurtosis of the XVZ returns shows that its performance is characterised by a few very large positive

returns. Timing an investment in XVZ would therefore be crucial, assuming one could always find a

buyer in the secondary market when wishing to ooad the investment. The simpler static strategy

seems a better bet for a longer-term investment, being much less exposed to extreme daily shocks.

Nevertheless, even the XVIX has a higher kurtosis than the tracker ETNs, so potential investors

should be aware that both products are exposed to significant daily fluctuations.

Panel B of Table 6 shows the correlation matrix of daily returns on the ETN2 with the daily

returns on the VIX and the VXV. The XVZ has a significant positive correlation with both the

VIX and the VXV (not surprizingly, since they are used as indicators for changing allocations)

but the static XVIX strategy has virtually zero correlation with the VIX and the VXV, indeed

the correlations are even slightly negative. This shows that the XVIX could be regarded as an

investment instrument which lies outside the equity volatility asset class.

4.3. Introducing Volatility ETN3

The XVIX and XVZ indicative returns between March 2004 and December 2011 are depicted

in the middle section of Fig. 6 below, along with some summary statistics. Later on, a more

detailed analysis will show that the XVIX and XVZ have a complementary performance. That is,

the XVIX performs best when the market is in contango and the XVZ only performs well during

market crashes, i.e. when the VIX futures term structure swings into steep backwardation, at

which time it performs very well indeed, exactly when the XVIX makes a significant loss. Hence,

we ask: it is possible to enhance returns further by holding a combination of the XVIX and XVZ?

To illustrate the possible benefits of such an ETN based on ETN2 (called an ETN3, for obvious

reasons) we now consider one static and one dynamic allocation to the two already-issued ETN2:

The static ETN3, CVIX, allocates 75% of capital to XVIX and 25% to XVZ. This allocation

is chosen because it corresponds almost exactly to the proportion of days in our sample that

the relationship between the 30-day contract and 150-day contract was in contango (75%)

and backwardation (25%), as previously noted;

The dynamic ETN3, CVZ, holds XVIX when the VIX term structure is in contango (i.e.

Yt < 1) and XVZ when VIX term structure is in backwardation (Yt 1).

Any prospective issuer of these ETN3s need not replicate the product by actually holding shares

of the XVIX and XVZ indeed, the ETN redemption constraints would preclude this. Instead,

direct trading along the term structure of VIX futures, with daily rebalancing to create synthetic

ICM futures as described in Section 2, allows their performance to be replicated exactly. Since

18

replication is so easy there are numerous other rules for allocating between the two ETN2 that

might be considered. Indeed, we expect major volatility ETN issuers to be actively developing

such products at the time of writing. The allocations specified above were merely the simplest

possible static and dynamic rules based on the observation that the XVIX tends to performs best

when the VIX term structure is in contango, and the VXZ performs better when the VIX term

structure is in backwardation.

Fig. 6 shows the daily returns to the VXX and VXZ (above), the XVIX and XVZ (middle)

and CVIX and CVZ (below) plus some descriptive statistics: the annualized mean daily return,

the volatility, Sharpe ratio and total return computed over the entire sample period, net of fees.

Clearly, the two VIX futures trackers VXX and VXZ are high-risk, low-return ETNs. They are

clearly undesirable as stand-alone investments and should only be used for short-term speculation.

By contrast, the differential roll-yield trades, XVIX and XVZ, have a much better risk-adjusted

performance: the XVIX has a lower risk than the XVZ, and a higher total return over the period.

However, its average daily return is lower, at only 15.94% in annual terms compared with 17.75%

for the XVZ.

Jan05Jan06Jan07Jan08Jan09Jan10Jan11Jan120.2

0

0.2

Mean = 43.37%, Vol = 56.69%, SR = 0.77, TR = 99.16%

VX

X r

etu

rns


0

0.2

Mean = 4.83%, Vol = 30.01%, SR = 0.16, TR = 52.80%

VX

Z r

etu

rns


0

0.2

Mean = 16.45%, Vol = 13.55%, SR = 1.21, TR = 250.28%

XV

IX r

etu

rns


0

0.2

Mean = 17.45%, Vol = 23.70%, SR = 0.74, TR = 227.08%

XV

Z r

etu

rns


0.1

0

0.1

0.2

Mean = 16.63%, Vol = 13.77%, SR = 1.21, TR = 254.48%

CV

IX r

etur

ns


0

0.2

Mean = 30.85%, Vol = 23.31%, SR = 1.32, TR = 870.57%

CV

Z re

turn

s

Fig. 6 Daily returns on volatility ETNs (VXX, VXZ, XVIX, XVZ, CVIX and CVZ), annualized mean daily return,volatility, Sharpe ratio and total return between March 26, 2004 and December 30, 2011.

19

The proposed ETN3 products, the CVIX and CVZ, are highly correlated with the two ETN2s:

the static products (XVIX and CVIX) have a sample correlation of 91.78% and the dynamic

products (XVZ and CVZ) have a sample correlation of 91.68%. The CVIX has a performance very

similar to the XVIX but the CVZ performs very much better than the XVZ. In fact, since 2004

CVZ has had an average daily return of over 30% (annualized), with a Sharpe ratio of 1.31 and a

total return of almost 800%.

4.4. Diversifying Risk along the VIX Futures Term Structure

Although the main objective when structuring an ETN product is to maximize return, there is

also a diversification benefit arising from any long-short position on VIX futures of two different

maturities, since their returns are highly correlated. This diversification effect could be further

enhanced by trading more than two different constant-maturity tracker ETNs. In this section

we consider how an ETN2 product might be structured by replicating a portfolio containing VIX

futures of all available maturities. Such portfolios should also be of internal interest to the issuing

bank. This is because the investment arm of a large commercial bank could be actively participating

in the VIX futures market in several ways other than issuing ETNs: (1) it may act as a market

maker or a broker executing clients orders;25 (2) other departments may use VIX futures to hedge

their volatility exposure; and (3) a separate proprietary trading desk might speculate in the market.

Since the bank may be actively engaged in all these operations simultaneously, open positions in

the VIX futures book of the bank overnight may be unavoidable. Thus, the risk management

department of the bank as a whole and its Chief Risk Officer in particular should be interested in

minimizing the risk exposure by effectively diversifying its positions along the VIX term structure.

Suppose that the bank has trades along the whole VIX futures term structure and is net long,

with weights vector wt = (w1t, w2t, . . . , wnt) where w1t + w2t + + wnt = 1. Denote by t theestimate of the n n covariance matrix of the returns on the contracts at time t. Then the globalminimum-variance portfolio, i.e. the exposure along the term structure that will give the lowest

variance at time t, is given by the weights vector:

wt =1t 1

11t 1(12)

where 1 is the n 1 vector of 1s. The minimum variance attainable is

2t = wt tw

t . (13)

Many banks would seek to be net long on VIX futures, not only to hedge the short volatility

exposure of writing options and variance swaps but also to accommodate the long volatility positions

of institutional clients that seek to diversify their equity exposures. However, a minimum-variance

portfolio of VIX futures can also be net short, with weights wt .25A list of authorised market makers and brokerage firms in VIX futures market can be found in http://cfe.

cboe.com/tradecfe/brokers.aspx.

20

We shall include the ETN derived from out-of-sample minimum variance allocations in our

empirical results below, and label this potential ETN2 product the MVX, for obvious reasons. For

this the estimate t of the covariance matrix in (13) is computed daily, using a simple rolling

equally-weighted historical covariance matrix calculated from the last N returns.26

5. Performance of ETNs: The Investors Perspective

This section reports detailed results on the comparative performance of VIX futures ETNs,

based on their indicative values between March 2004 and December 2011. First we give a brief

description of the performance criteria that we apply.

5.1. Performance Measures

The most widely used risk-adjusted performance measure, introduced by Sharpe (1966, 1975),

is the ratio of the expected excess annual return on a portfolio to its standard deviation. We follow

standard practice by estimating the Sharpe ratio as the mean of daily excess returns divided by its

standard deviation, annualizing this ratio through multiplying by250. The Sharpe ratio adjusts

for risk as measured by the portfolio volatility whereas the similar performance measure introduced

by Sortino and Van Der Meer (1991) adjusts only for downside risk, as measured by the square root

of semi-variance introduced by Markowitz (1959). In addition to the Sharpe and Sortino ratios,

we examine the Omega ratio introduced by Keating and Shadwick (2002), which is the ratio of

the expectation of the positive excess returns to the expectation of the negative excess returns.

A positive Omega ratio indicates that positive returns tend to outweigh the negative returns, on

average. As another measure of downside risk which quantifies the extent of possible margin calls

on the issuer of the ETN, we calculate the maximum daily drawdown over the entire sample from

March 2004 to September 2011.

Finally we compute the manipulation-proof performance measure Theta derived by Goetzmann,

Ingersoll, Spiegel, and Welch (2007). Manipulation of a performance measure is due to the fund

managers actions, like window dressing or closet indexing, which increase the value of a performance

measure, but add no real value to the investor. The properties of such a manipulation-proof

performance measure are similar to a power utility pay-off. In particular, the measure depends on

the relative risk aversion of the investor, . We set = 3, but our empirical results are qualitatively

robust to the choice of this parameter. For a portfolio with daily excess returns rt measured over

a sample size of size T it is defined as

=250

(1 ) ln(T1

Tt=1

(1 + rt)1

). (14)

Similar to Ederington (1979) the effectiveness of the diversification is measured by the per-

26But the disadvantages of this approach are well known - see Alexander (2009) for example. Baba, Engle, Kraft,and Kroner (1991) and many others advocate instead the multivariate GARCH model, where a conditional covariancematrix is estimated at every time t using maximum likelihood methods. The results of using a GARCH covariancematrix are given in the appendix.

21

centage reduction in variance when the minimum-variance portfolio is held, relative to holding the

undiversified benchmark position over a period of one day. That is,

Et =2,t 2pi,t

2,t, where (15)

2,t = (1 )i=1

i1(rti

)2, and 2pi,t = (1 )

i=1

i1(rpiti

)2, (16)

where rt and rpit are the daily returns on the benchmark portfolio, and on the minimum-variance

portfolio, respectively. Here 0 1, with higher values resulting in smoother time series for2,t and

2pi,t. When = 1 we have the standard equally-weighted sample variances, but we prefer

to use < 1 for exponentially smoothed sample variances as these change over time according to

market conditions.

5.2. Empirical Performance of ETNs

Table 7 reports the range of performance measures described above, applied to the already-

issued ETNs in question (i.e. the VXX, VXZ, XVIX, and XVZ) over the entire sample period and

over three sub-periods: March 26, 2004 October 31, 2006 (a period of stable trending equity

markets); November 1, 2006 May 31, 2009 (which covers the credit and banking crises); and June

1, 2009 December 30, 2011 (a period that includes the Eurozone debt crisis). In each period we

compare their performance to the performance of the ETN3 products defined above, i.e. the CVIX

and CVZ. Panel A results for the tracker ETNs show that long-term investments in these products

are unwise. Although the VXZ performs better than the VXX, the VXZ could still incur a daily

loss of around 10%, and the expected return is negative. The only sub-period when the trackerETNs performed reasonably well was the central period covering the credit and banking crises, and

over this period the long-term VIX futures tracker VXZ outperformed the short-term tracker VXX

according to each criterion.

Turning now to the ETN2s: both XVIX and XVZ outperform the tracker ETNs over the entire

period, and over the first and last sub-samples in Panel B and Panel D (when the XVIX also

outperforms the XVZ). This is according to each criterion, except for the large maximum daily loss

observed for the XVZ. During the credit and banking crisis period (Panel C) their performance was

comparable to that of the tracker ETNs, with the XVZ clearly a better investment than the XVIX.

Still, the large daily drawdowns highlight that the VIX futures market can swing very quickly from

contango into strong backwardation, with inverse contracts rapidly appreciating in value, especially

the short-term ones. In this case the potential profits from differential roll-yield strategies can easily

turn into large losses.

The performance of the ETNs and ETN2s is clearly regime-dependant, and the aim of our

proposed ETN3s is to build portfolios that will perform relatively well irrespective of the market

regime or at least be more immune to the regime than the already-issued ETNs. Examining the

performance of the ETN3s over the whole period, both the static mix of XVIX and XVZ (i.e. the

22

Table 7

Performance measures for ETNn, n = 1, 2, 3.

ETN ETN2 ETN3

VXX VXZ XVIX XVZ CVIX CVZ

Panel A: Full SampleSharpe ratio -0.6328 -0.0498 1.1875 0.7394 1.1896 1.3111Sortino ratio -0.9789 -0.0761 1.7322 1.2106 1.7893 2.1649Omega ratio 0.8924 0.9909 1.2315 1.1858 1.2430 1.3424Maximum daily loss -16.35% -9.16% -7.09% -10.96% -6.25% -10.96%Theta -81.69% -14.85% 13.22% 9.33% 13.50% 22.71%

Panel B: March 26, 2004 - October 31, 2006Sharpe ratio -2.6751 -1.5779 1.2081 -0.3174 0.8276 0.7343Sortino ratio -3.3698 -2.0236 1.7334 -0.4047 1.1418 0.9635Omega ratio 0.6249 0.7537 1.2320 0.9341 1.1576 1.1563Maximum daily loss -16.35% -9.16% -3.65% -10.96% -3.96% -10.96%Theta -112.47% -37.89% 12.45% -8.06% 7.66% 7.64%

Panel C: November 1, 2006 - May 31, 2009Sharpe ratio 0.7697 1.0588 0.7664 1.3681 1.1579 1.5526Sortino ratio 1.2706 1.7141 1.1110 2.3175 1.7958 2.6325Omega ratio 1.1416 1.2109 1.1451 1.3279 1.2368 1.3851Maximum daily loss -12.58% -7.43% -7.09% -8.05% -6.25% -8.05%Theta -8.83% 19.04% 8.59% 28.79% 15.77% 34.08%

Panel D: June 1, 2009 - December 30, 2011Sharpe ratio -0.9295 -0.3008 1.8812 0.6555 1.7468 1.5454Sortino ratio -1.4563 -0.4668 2.8551 1.1612 2.7270 2.8449Omega ratio 0.8483 0.9476 1.3662 1.1762 1.3516 1.4558Maximum daily loss -16.12% -8.64% -2.32% -9.65% -2.52% -9.65%Theta -122.63% -25.24% 18.58% 7.51% 17.09% 26.56%

Table reports various performance measures for the excess returns, net of fees, on the two tracker ETNs (VXXand VXZ) the two ETN2 differential roll-yield strategies (XVIX and XVZ) and their static and dynamic ETN3

combinations (CVIX and CVZ). The full sample is between March 26, 2004 and December 30, 2011.

CVIX) and the dynamic, regime-dependant mix (i.e. the CVZ) have higher Sharpe ratios, Sortino

ratios, Omega ratios and Theta than any of the existing ETNs. However, the XVIX outperforms

both ETN3s during the first sub-sample and XVZ outperforms CVIX during the second sub-sample.

CVZ experiences the same large drawdowns as the XVZ but they are still much smaller than those

experienced on the VXX. According to the ultimate test of the manipulation-proof measure Theta,

the CVZ is the preferred amongst all the ETNs except during the first sub-sample, when the XVIX

performed best. As with the two ETN2s, the static CVIX outperforms the dynamic CVZ only

during the stable trending market of the first sub-sample.

5.3. The Minimum-Variance ETN

Fig. 7 depicts the minimum-variance weights derived using Eq. (12), starting in March 28, 2005.

They are fairly stable over time, generally in the range of [1, 1], except for the long-term contractsthat were trading on very low volumes until mid 2007. As expected, the minimum-variance portfolio

is balanced, with long and short positions evenly distributed.

VIX futures trading is very strongly concentrated on contracts that are relatively close to

23

Jan06 Jan07 Jan08 Jan09 Jan10 Jan113

2

1

0

1

2

3

4

Minim

um Va

rianc

e Weig

hts


Fig. 7 Minimum variance weights on the constant-maturity VIX futures based on equally weighted covariancematrices computed using a rolling-window methodology.

maturity. For instance, during the first two weeks of August 2011 the average daily volume of VIX

futures traded was about 40,000 on the August contract, 34,000 on the September contract, 10,000

on the October contract and then decreasing from about 7,500 on the November contract to less

than a 1,000 contracts per day for the February and March 2012 expires. This was a volatile period

and overnight risks might be considerably reduced by taking a more balanced exposure across the

whole term structure, as prescribed by the minimum-variance portfolio, MVX.

On each day between March 28, 2005 and December 30, 2011 we compute the equally weighted

covariance matrix t on the ICM VIX futures contracts, based on the last 250 returns.27 The

minimum-variance positions are kept for one day and the realized out-of-sample portfolio return

is recorded. Then we roll the sample forward one day and repeat to obtain a series {rpit }Tt=N+1 ofout-of-sample returns on the minimum-variance portfolio.

We label the ETN based on minimum variance portfolio weights the MVX. Indicative returns

on the MVX can only be computed from March 2005, since we have (arbitrarily) employed a 1-year

sample to estimate the required covariance matrix. Over the period March 2005 December 2011

the MVX had a Sharpe ratio of 0.89 with a volatility of 23.58%, not including investor fees. Over

the same period the volatility of the CVZ was only marginally higher (24.61%) and the CVIX had

a much lower volatility (13.90%). This is possible because the minimum-variance allocation is only

in-sample, and the returns are recorded out-of-sample.

We now consider the diversification effectiveness of the MVX relative to a benchmark portfolio

consisting of the VXX. That is, we set = 30 days in Eqs. (15) and (16) and use the RiskMetrics

27We also used several other values for N , but the variance reduction was fairly robust to this choice. Results areavailable from the authors on request.

24

Jan06 Jan07 Jan08 Jan09 Jan10 Jan1120%

30%

40%

50%

60%

70%

80%

90%

100%

Varia

nce R

educ

tion (

E t)

MVX XVIX XVZ CVIX CVZ

Fig. 8 The variance reduction from exposure to MVX, or one of the new static or dynamic differential roll-yieldETNs, relative to an exposure to VXX. A figure of 90% indicates that the variance of the MVX (or other ETN) is10% of the variance of the VXX.

value = 0.94 in Eq. (16) and hence compute the time series Et of variance reducation. For

comparison, we also replace the variance of MVX by the variance of XVIX, of XVZ, CVIX and

CVZ, yielding the five different time series of variance reduction effectiveness displayed in Fig. 8.

Surprizingly, a the greatest variance reduction is achieved with the static ETN2 and ETN3, i.e. the

XVIX and CVIX. These ETNs typically have variance between 5% and 20% of the VXX, in other

words the variance reduction is between 95% and 80%. During the last two years, their variance has

never exceeded 10% of the variance of VXX. This shows that it is difficult to construct allocations

along the VIX term structure that have a lower volatility than the existing (and potential) static

differential roll-yield ETNs.

From the issuers perspective, the MVX has some advantages over single tracker products, which

are hedged using only two VIX futures contracts. Concentrated exposures often incur margin calls,

especially during volatile periods, but margin costs for a diversified portfolio should be very small

because the exposure is balanced: brokers typically charge only a few basis points on the difference

between interest earned on the margin when the futures price rises, and interest paid on the margin

when the futures price falls. On the other hand, diversifying a concentrated exposure along the term

structure incurs a hedging cost all along the curve. Apart from considering the internal activities

of the bank, the attraction of issuing a diversified ETN such as MVX will depend on the relative

size of fees and hedging costs plus costs of maintaining a margin account.

5.4. Risk and Return on ETN2 and ETN3

Here we present a detailed analysis of the risk-returns profile of all the ETN2 and ETN3 prod-

ucts considered so far. Fig. 9 presents a rolling picture of the Sharpe ratio, mean daily return

25

(annualized) and the annual volatility of the ETN2 and ETN3 products, including the minimum-

variance ETN (MVX). Each statistic is computed using a sample of the previous 250 days, then

the sample is rolled over daily and the statistics are re-calculated.

Jan06 Jan07 Jan08 Jan09 Jan10 Jan112.00

0

2.00

4.00

6.00

Shar

pe R

atio

Jan06 Jan07 Jan08 Jan09 Jan10 Jan1140%

0%

40%

80%

120%

Mea

n Re

turn

Jan06 Jan07 Jan08 Jan09 Jan10 Jan110

12.5%

25%

37.5%

50%

Vola

tility

XVIX XVZ CVIX CVZ MVX

Fig. 9 Sharpe ratio, mean and volatility computed ex-post using the previous 250 days of returns on the ETN2 andETN3 and on the minimum-variance ETN (MVX), March 2005 December 2011.

Whilst the dynamic products XVZ and CVZ have a consistently higher volatility, they also have

an extremely large return in Q4 2008, at the onset of the banking crisis. Surprizingly, the MVX

also has a very good return indeed it produced the best performance of all at the end of 2007

and in Q2 2010. The two static ETNs, the XVIX and CVIX, have the lowest volatility throughout

and a consistently positive mean return until the end of the sample. Unfortunately, during 2011

these static strategies performed very badly indeed, with a Sharpe ratio rapidly falling from an

unprecedented high of almost 5 (!) into negative territory within the space of 12 months. The

Sharpe ratio of all ETNs fell during this period, and the CVZ it is the only one of the seven ETNs

that consistently held a positive mean return, when computed over the previous 250 days, over the

entire sample period.

5.5. Comparison with Standard Asset Classes

Next we compare the performance of the proposed ETN3 products with the performance of

standard asset classes (US equities, commodities and bonds) and assess the potential for diversi-

fication of investments in standard assets by allocating a portion of capital to the CVIX or the

26

CVZ. To proxy the US equity market we use the S&P 500 stock index; to proxy the US commodity

market we employ the S&P GSCI commodity index (SPGSCI) which allocates between the major

physical commodities that have active futures traded on US exchanges; and as a proxy for the US

fixed income market we use the Barclays Aggregate Bond Fund (AGG) which corresponds to the

performance of the total US investment grade bond market. We subtract a proxy for the risk-free

rate (91-day U.S. T-bill) from the returns on all variables to construct excess over the risk-free rate

returns. Again, we use the longest possible historical period (almost 8 years) for this analysis.

Table 8

Descriptive statistics and correlation matrix of ETN3s and traditional asset classes.

Panel A: Summary statisticsAnnualized Mean Volatility Sharpe Ratio Skewness Kurtosis Total Return

CVIX 16.32% 13.72% 1.1896 0.15 6.68 232.95%CVZ 30.80% 23.49% 1.3111 1.48 18.92 798.35%S&P 500 2.75% 22.17% 0.1239 -0.04 9.93 2.28%SPGSCI 3.68% 5.77% 0.6383 -2.60 73.50 31.67%AGG 13.05% 26.90% 0.4850 -0.14 2.03 109.02%

Panel B: CorrelationsCVIX CVZ S&P 500 SPGSCI AGG VIX

CVIX 1.00CVZ 0.80 1.00S&P 500 -0.35 -0.53 1.00SPGSCI -0.11 -0.17 0.33 1.00AGG 0.11 0.04 -0.12 0.01 1.00VIX 0.24 0.40 -0.75 -0.24 0.11 1.00

Panel A reports the summary statistics for daily excess returns on the two volatility ETN3 products that we propose,and compares them with those for the S&P 500, the SPGSCI commodity index and the AGG bond index. Samplemeans, standard deviations and corresponding Sharpe ratios are annualized based on 250 trading days per year.Panel B is the correlation matrix for these returns and the VIX. The sample is March 26, 2004 to December 30, 2011.

Table 8 reports the results. Panel A shows that the ETN3 products clearly outperformed the

three standard asset classes. Investing in commodities was profitable but highly risky and the

Sharpe ratio for commodities is much worse than those of the ETN3s. Results are even worse for

the other two asset classes; investing in equity would have been only marginally profitable with an

annualized mean return of just 2.75%. Moreover, if a fund manager had employed a buy-and-hold

strategy over the whole period he would have experienced a return of just 2.28%. Investment in

the bond market would have yielded the highest total return of the three conventional asset classes,

but it is still far less than the return that could have been realized had the CVIX and CVZ been

available to investors. Total returns from buy-and-hold from March 2004 to December 2011 were

almost 235% for the CVIX and almost 800% for the CVZ. The CVIX has negligible skewness and

relatively low kurtosis. Whilst the CVZ has a high kurtosis (almost 19) it also has a strong positive

skewness, so extreme positive returns have been more common than extreme negative returns.

A considerable potential for diversification with standard asset classes is also offered by the

CVIX, and even more so by the CVZ. Panel B of Table 8 shows that both ETNs have a weak

positive correlation with the VIX index and a strong negative correlation with the S&P 500 equity

index. They also display a small negative correlation with the SPGSCI commodity index and the

CVIX has a small positive correlation with the AGG bond index. By comparison, the correlation

27

between the S&P 500 and the SPGSCI indices is, at 0.33, relatively strong. Moreover, the CVZ

has a correlation with the AGG bond index that is insignificantly different from zero.

6. Hedging ETNs: Concerns for Issuers and Regulators

The holder of an ETN has the right to redeem early (typically in lots of 25,000 shares) at the

closing indicative value one business day after the holder gives notice (through his broker) of the

redemption. During volatile periods early redemptions can be extremely large. For instance, during

the first few days of the Euro crisis in August 2011 the number of VXX shares outstanding more

or less halved, from 42 million to 21 million. The early redemption features of ETNs require the

issuer to secure a profit through hedging his exposure daily, to earn at least the annual service fee

after accounting for hedging costs.

In this section we derive expressions for net assets and liabilities on the hedging account; we illus-

trate the extent of profits that can be made by banks that issue ETNs, using some straightforward

examples; we perform a scenario analysis, which issuers should be employing for risk management;

and explain to regulators the mechanics (including various avenues for front-running) of the hedging

activities that tie the NYSE with the CBOE.

6.1. Perfect Hedges Based on Indicative Values

Let xmt denote the net liability facing the issuer of a single constant-maturity tracker ETN, with

maturity m, on day t. Let rmt+1 denote the return on the m-maturity S&P index from day t to day

t+ 1, which is given by Eq. (2). Let fm denote the (daily equivalent) constant investment fee and

let r denote the daily return on the risk-free asset. Then from day t to day t + 1 the liability xmt

changes to

xmt+1 = xmt (1 + r

mt+1 + r fm).

The assets held by the issuer in the risk-free asset are xm0 on the day of issue and thereafter the

assets ymt on day t grow at the risk-free rate r, plus the P&L from the hedge of his liabilities on

day t. Suppose the issuer hedges his liabilities xmt at time t using the synthetic m-maturity futures.

Ignoring any margin costs, taking an exposure of xmt in the futures with return rmt+1 adds the P&L

xmt rmt+1 to the issuers assets, so that in the absence of transaction costs his total assets at time

t+ 1 would be given by ymt (1 + r) + xmt r

mt+1. However, there is a small daily transaction cost of

hmt+1 =smt+12|xmt + 1 xmt (1 + rmt+1)|

arising from rebalancing the positions in the two futures that straddle the target maturity.28 Here

28Recall that during a weekday the amount (l s)1, which is usually 1/28 or 1/32 of the exposure, is transferredfrom the shorter to the longer maturity futures. Over a weekend it is 3 times this amount.

28

smt represents the bid-ask spread for an m-maturity futures exposure on day t.29 Hence,

ymt+1 = ymt (1 + r) + x

mt r

mt+1 hmt+1.

Thus, the net assets (i.e. assets liabilities) at time t+ 1 are given by

zmt+1 = ymt+1 xmt+1 = (ymt xmt )(1 + r) + xmt fm hmt+1. (17)

6.2. Empirical Examples

Fig. 10 depicts three empirical examples showing, on the left, the daily average flow of fees and

hedging costs, averaged over the previous month. The right-hand graphs depict the cumulative

net assets accruing to the issuer of ETNs, as a proportion of the current liability. The examples

are: (i) an initial issue of equal amounts in the 30-day tracker VXX and its inverse tracker, the

AAVX; (ii) an initial issue of XVIX, plus two-thirds of this exposure in the inverse 150-day tracker

and one-third in the direct 30-day tracker; and (iii) an initial issue of equal amounts on the XVIX,

and the four direct and inverse trackers at 30-day and 150-day maturities, i.e. five ETNs in all.

We suppose all shares in issue were sold on the primary market on March 30, 2004 and that the

number of shares outstanding is assumed constant, i.e. there are number issues of shares less the

shares redeemed early on any day is assumed constant.

The net asset path (assets minus liabilities) is not shown, but it grows monotonically over the

entire period in all three cases. Net assets as a proportion of the current liabilities also remains

positive over the entire period even though the fees earned cover the hedging costs only in case

(i) and (iii). In case (ii) the excess of hedging costs over fees earned erodes the asset side of the

balance sheet so that they are much less than in cases (i) and (iii).

Of course, the initial exposures change over time. In particular, the liabilities for direct and

inverse trackers vary inversely, so that the liability to the inverse tracker grows when the liability to

the direct tracker falls. For this reason the net assets can fall, as a proportion of the liabilities. For

instance, by March 20, 2009 the net assets in case (i) i.e. an initial issue of equal dollar amounts

in VXX and AAVX had risen to nearly 85% of the current liability, net of fees and hedging

costs. But one year later, by March 20, 2010 it had fallen steadily too less than 30% of the current

liability. This is because the liability to the inverse tracker was gradually rising during the long

contango market following the banking crisis. Clearly the issuer of a suite of products may find it

advantageous to issue further notes in direct tracker products at such times, if possible. This way,

the basket of ETNs can be kept in balance so that not only the net assets are always growing, but

also net assets are always growing as a proportion of the current liability.

29In the next section we report an empirical analysis of net assets accruing to the issuer of ETNs over time,reporting fees and hedging costs separately. Here we assume constant spreads of 25bps for the 30-day and 30bps forthe 150-day trackers. It is true that spreads have decreased dramatically over time, but our purpose is to assess theprofitability of a potential future issuance of ETNs, so it is appropriate to assume hedging costs are fixed at theircurrent size.

29

Jan05 Jan06 Jan07 Jan08 Jan09 Jan10 Jan110

5

10

15Long tracker and inverse


20%

40%

60%

80%

100%Long tracker and inverse


0.5

1

1.5

2Suite of three

Understanding ETNs on VIX Futures

Documents

Transcript of Understanding ETNs on VIX Futures